Multi-Object 3D Grounding with Dynamic Modules and Language-Informed Spatial Attention